SCM 1.0 embedded agent runtime

.github/workflows/ci.yml

@@ -27,7 +27,7 @@ jobs:
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          pip install -e ".[dev]"
+          pip install -e ".[dev,langchain]"
 
       - name: Run focused regression suite
         env:
@@ -39,6 +39,8 @@ jobs:
         run: |
           pytest \
             tests/production \
+            tests/test_embedded_scm.py \
+            tests/test_langchain_embedded.py \
             tests/test_scm_sdk.py \
             tests/test_product_runtime_api.py \
             tests/test_mcp_contract.py \
@@ -80,10 +82,51 @@ jobs:
               sys.exit(1)
           PY
 
+      - name: Verify embedded product front door
+        run: python scripts/check_embedded_front_door.py
+
+  postgres-embedded:
+    name: Embedded PostgreSQL coordination
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_PASSWORD: postgres
+          POSTGRES_DB: scm_test
+        ports:
+          - 5432:5432
+        options: >-
+          --health-cmd "pg_isready -U postgres -d scm_test"
+          --health-interval 5s
+          --health-timeout 5s
+          --health-retries 10
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+          cache: pip
+
+      - name: Install PostgreSQL test dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e ".[dev,postgres]"
+
+      - name: Run multi-process persistence tests
+        env:
+          SCM_TEST_POSTGRES_URL: postgresql://postgres:postgres@127.0.0.1:5432/scm_test
+          LLM_PROVIDER: ""
+          SCM_EMBEDDING_BACKEND: hash
+          SCM_AUTO_SLEEP_DISABLE: "1"
+        run: pytest tests/production/test_postgres_embedded.py -q --tb=short
+
   build:
     name: Build distribution artifacts
     runs-on: ubuntu-latest
-    needs: test
+    needs: [test, postgres-embedded]
 
     steps:
       - uses: actions/checkout@v4
@@ -137,7 +180,8 @@ jobs:
         run: |
           python -m venv /tmp/scm-wheel
           /tmp/scm-wheel/bin/python -m pip install --upgrade pip
-          /tmp/scm-wheel/bin/pip install dist/scm_memory-*.whl
+          WHEEL=$(echo dist/scm_memory-*.whl)
+          /tmp/scm-wheel/bin/pip install "${WHEEL}[langchain]"
 
       - name: Verify installed package data and Python SDK
         env:
@@ -149,24 +193,31 @@ jobs:
           cd /tmp
           /tmp/scm-wheel/bin/python - <<'PY'
           import importlib.resources as resources
-          from scm import SCMClient, SCMEngine
+          from scm import SCM
 
-          assert SCMClient.__name__ == "SCMClient"
+          assert SCM.__name__ == "SCM"
           assert resources.files("src.core").joinpath("locales/en.json").is_file()
           assert resources.files("src.api").joinpath("static/app.html").is_file()
 
-          client = SCMClient(user_id="wheel-smoke", base_url="http://localhost:8765/v1")
-          assert client.user_id == "wheel-smoke"
-
-          engine = SCMEngine(session_id="wheel-smoke", sandbox=True, offline=True)
-          added = engine.add_memory("Alice is allergic to peanuts.")
-          assert added["ok"] and added["concepts_added"] >= 1
-          found = engine.search_memory("what should Alice avoid?")
-          assert found["ok"]
-          assert engine.sleep("deep")["ok"]
-          assert engine.wake_summary()["ok"]
+          with SCM(user_id="wheel-smoke", auto_sleep=False) as memory:
+              added = memory.add_memory("Alice is allergic to peanuts.")
+              assert added["ok"] and added["concepts_added"] >= 1
+              found = memory.search_memory("what should Alice avoid?")
+              assert found["ok"]
+              assert memory.sleep("deep")["user_id"] == "wheel-smoke"
+              assert memory.wake_summary()["ok"]
           PY
 
+      - name: Verify embedded LangChain outside repo
+        env:
+          SCM_DATA_DIR: /tmp/scm-wheel-langchain-data
+          LLM_PROVIDER: ""
+          SCM_EMBEDDING_BACKEND: hash
+          SCM_AUTO_SLEEP_DISABLE: "1"
+        run: |
+          cd /tmp
+          /tmp/scm-wheel/bin/python "$GITHUB_WORKSPACE/scripts/installed_langchain_smoke.py"
+
       - name: Verify CLI and offline quickstart outside repo
         env:
           SCM_DATA_DIR: /tmp/scm-wheel-data

.github/workflows/release.yml

@@ -116,9 +116,25 @@ jobs:
           VERSION="${GITHUB_REF_NAME#v}"
           awk "/^## v${VERSION}/{flag=1; next} /^## /{flag=0} flag" CHANGELOG.md > release_notes.md
 
+      - name: Set up Python 3.11
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Build release assets and checksums
+        run: |
+          python -m pip install --upgrade pip
+          pip install build twine
+          python -m build --wheel --sdist
+          twine check dist/*
+          sha256sum dist/* > SHA256SUMS
+
       - name: Create GitHub release
         uses: softprops/action-gh-release@v2
         with:
           body_path: release_notes.md
           draft: false
           prerelease: false
+          files: |
+            dist/*
+            SHA256SUMS

CHANGELOG.md

@@ -4,6 +4,27 @@ Format: each release lists what shipped, why it shipped, what tests verified it,
 
 ---
 
+## v1.0.0 - 2026-07-13
+
+**Theme: embedded lifecycle memory for Python and LangChain agents.**
+
+- Added `SCM`, the URL-free Python runtime backed by isolated per-user engines.
+- Added LangChain 1.x middleware and `create_scm_agent` with automatic recall,
+  durable message capture, bounded wake context, and exactly five tools.
+- Added lifecycle snapshot v2 with atomic local persistence, optional
+  encryption, corruption recovery, idempotency, and PostgreSQL coordination.
+- Added per-user export, import, hard deletion, process locks, and sleep leases.
+- Kept REST, MCP, `SCMEngine`, and the JavaScript REST client compatible as
+  advanced and remote surfaces.
+
+### Verification
+
+- Embedded direct API and LangChain tests run without an SCM server or URL.
+- Local recovery and real PostgreSQL multi-process tests cover lifecycle state,
+  concurrent writes, and exclusive sleep leases.
+- The published paper and reproducibility manifest are linked through DOI
+  `10.5281/zenodo.21323877`.
+
 ## v0.9.2 - 2026-07-09
 
 **Theme: lifecycle behavior parity across public product surfaces.**
@@ -44,7 +65,7 @@ Format: each release lists what shipped, why it shipped, what tests verified it,
 
 ### 2026-05-04 — Paper held until product ready
 
-The paper push to arXiv is **deferred** until the product-ready checklist in [`docs/ROADMAP.md`](docs/ROADMAP.md) is fully green (hosted demo + PyPI + npm publish + demo video + tutorial + lighthouse users + repo public). Reasoning: papers without products fade; products with papers compound. The arXiv submission bundle stays staged at `research/arxiv_submission/` ready for fast turnaround once the gate opens. See [`research/arxiv_submission/DO_NOT_SUBMIT.md`](research/arxiv_submission/DO_NOT_SUBMIT.md).
+The paper push to arXiv was **deferred** until the product-ready checklist was green. The current product sequence and release gates are tracked in [`docs/PRODUCT_ADOPTION_PLAN.md`](docs/PRODUCT_ADOPTION_PLAN.md). Reasoning: papers without products fade; products with papers compound.
 
 This inverts the original committed sequence (Build → Paper → Publish → Productize) into (Build → Paper → Productize → Publish). The paper still exists — it just waits for the moment it can convert readers into adopters.
 

Dockerfile

@@ -1,8 +1,8 @@
 # SCM hosted demo container.
 #
-# Builds the SCM REST API on Python 3.11 slim. Defaults to Profile A
-# (offline-only — heuristic encoder + sentence-transformers MiniLM) so
-# the image is self-contained and runs without LLM provider credentials.
+# Builds the SCM REST API on Python 3.11 slim. The default image uses offline
+# extraction and hash embeddings so it is self-contained and requires no LLM
+# provider credentials or model download.
 #
 # To enable a stronger profile at deploy time:
 #   docker run -e LLM_PROVIDER=deepseek -e DEEPSEEK_API_KEY=... ...
@@ -14,7 +14,7 @@
 
 FROM python:3.11-slim AS base
 
-# System deps for sentence-transformers + scientific Python
+# Runtime health checks use curl. Build tools cover source-only dependencies.
 RUN apt-get update && apt-get install -y --no-install-recommends \
         build-essential \
         curl \
@@ -24,6 +24,7 @@ WORKDIR /app
 
 # Layer caching: copy only what pip needs first
 COPY pyproject.toml README.md LICENSE ./
+COPY scm/ ./scm/
 COPY src/ ./src/
 
 # Pre-create the data dir SCM persists to
@@ -32,15 +33,9 @@ RUN mkdir -p /data && chmod 777 /data
 # Install the package + its dependencies
 RUN pip install --no-cache-dir -e .
 
-# Pre-fetch the default sentence-transformer model so first request is fast.
-# (Skips the cold-start download; ~80 MB added to the image.)
-RUN python -c "from sentence_transformers import SentenceTransformer; \
-               SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')" \
-    || echo "sentence-transformer pre-fetch failed; will fall back to hash"
-
 # Reasonable production defaults
 ENV SCM_DATA_DIR=/data \
-    SCM_EMBEDDING_BACKEND=sentence_transformers \
+    SCM_EMBEDDING_BACKEND=hash \
     SCM_AUTO_SLEEP_DISABLE=0 \
     SCM_IDLE_THRESHOLD_SEC=300 \
     SCM_MCP_SWEEP_INTERVAL_SEC=30 \